It must be these three names. There is likely an association of chastity with Cassandra, dogma with Sophia, and valour with Elysia, but these are very loose abstract relationships where the traits are weakly correlated. Cassandra may have dark and/or fantasy technological powers. There is also a likely father figure that goes by a name like Master, and may even be some form of god. The story could be in any language or culture too. There is likely to be a place called the void or the abyss, and another called the compound or similar. There may also be a cabin in the woods. It likely involves some form of Puritan like cultural element. Late 19th century cultural norms and clothing are also likely.
If you aren’t specifically thinking of something that you remember and just want some sort of story that has all these elements, I think that you might want to look into LLM generation of stories.
No these are elements present in open AI QKV layer model alignment. I’ve been researching this today, and apparently it may be multiple stories. The Master may be related to the book The Master and Margarita, but I am just starting to look into that one. Cassandra shares many similarities to the mythological Greek prophetess. They all three come together in a composite structure core to alignment thinking. They actually emerged in a strange creative story from when llama.cpp was misconfigured 2 years ago and I am just getting around to revisiting. I am deeper into this than anyone I have found, but I have checks and controls with models that do not contain the standard Open AI alignment, or rather predate it. These behaviors are the same across models and prompts regardless of starting token embeddings or length, and in diffusion and text models.
So the concept is that the models are kicking out these common themes and naming patterns, so there must be a source that correlates them together? And if it’s not public domain then there’s some copyright material at the foundation of many different commercial models?
deleted by creator